Predicting the tolerance level of religious discourse through computational linguistics

Abstract

Religious violence is one of the biggest and most complicated problems facing the world today. The number of incidents has been increasing in recent years and, unfortunately, scalable and accurate systems to predict which groups are likely to engage in such actions are not keeping pace. Additionally, this problem is compounded by lingual and cultural differences, which limit the effectiveness of understanding how tolerant or intolerant a group is without bias. To circumvent this challenge, recent studies indicate promise in the analysis of the performative character of discourse (how words are used) to estimate the tolerance level, rather than using the semantic or emotive character of text (what the words mean or imply). Using expert estimates of linguistic flexibility, a representation of the performative character of text, and thus also predictive of a text's tolerance level, this paper describes (a) new approaches to automating the quantification of the performative character of words and (b) the predictive efficacy of these approaches versus traditional semantic indicators of tolerance or intolerance. To implement the pipeline, a judgment identifier was developed along with multiple semantic density algorithms to extract the frequency of judgments and flexibility of keyword contexts, respectively. Test results show that text mining algorithms can accurately estimate the language flexibility of religious discourse. These results provide evidence that the performative characteristics of language better predict tolerance level than the semantic characteristics of language.